It's probably just me, but the video in these video calls is mostly a waste of bandwidth, I am never paying attention to the bobbing heads in the screen. My point being, I, for one, would not notice very quickly if the person speaking was just a loop of someone looking interested, I sort of want to do this myself, OBS is probably the key, when called to speak you could switch scenes to the live feed.

I did some research into using ffmpeg as a sort of live network based video router and compositor, My conclusion was that it would work. however my heart was set on actual transparent rgba video and support for that was finicky.